LazySEO › Blog › What Role Do Backlinks Play in AI Assistant Citation Decisions?

What Role Do Backlinks Play in AI Assistant Citation Decisions?

LazySEO · Jul 31, 2026

Backlinks help indirectly by supporting authority, discovery, and retrieval eligibility—but they do not directly determine whether an AI assistant cites a page. For AI visibility, relevant content, credible evidence, crawlability, brand recognition, and platform-specific retrieval behavior can matter as much as—or more than—link totals.

In this article, Generative Engine Optimization (GEO) means improving a brand’s visibility in AI-generated answers. GEO is related to SEO, but the outcomes are different: traditional SEO focuses largely on search rankings and clicks, while AI visibility includes retrieval eligibility, source selection, answer generation, brand mentions, and linked citations.

Do backlinks directly determine whether AI assistants cite a page?

No. A backlink is not a guaranteed vote for inclusion in an AI-generated answer. It may contribute to a site’s broader authority or search eligibility, but an AI system still needs to retrieve the page, determine that it is relevant to the prompt, assess whether it supports the answer, and decide whether to mention or link to it.

These stages should not be treated as interchangeable:

For Google AI Overviews and AI Mode, Google says the foundational requirements are the same as for Google Search. A page must be indexed and eligible to appear in Google Search with a snippet, although meeting those requirements does not guarantee crawling, indexing, or inclusion in a generative AI feature. Google also says there is no special AI-only markup requirement. (developers.google.com)

That guidance applies specifically to Google’s generative AI features, not to every AI assistant. ChatGPT, Gemini, and Perplexity may use different indexes, retrieval systems, prompt sets, and source-selection processes.

Does backlink quality matter more than quantity?

The available vendor studies generally point toward quality and relevance over raw link volume, but their results are correlational and should not be interpreted as proof that earning a link causes an AI citation.

What the Semrush study measured

A Semrush study published on October 16, 2025, examined 1,000 randomly selected domains from its AI Visibility Toolkit. It compared backlink-related metrics with two outcomes:

• AI mentions: how often a domain was referenced in AI-generated answers.
• AI Share of Voice: a domain’s visibility in AI-generated answers relative to non-branded queries.

The study tracked ChatGPT, ChatGPT with Search, Gemini, Google AI Overviews, and Perplexity. It used Pearson and Spearman correlations to compare those outcomes with backlink metrics, including total links, referring domains, follow versus nofollow links, image versus text links, and Semrush Authority Score. (semrush.com)

The follow and nofollow results are close, but they do not show that nofollow links pass conventional ranking equity. They show only that, in this dataset, the number of follow or nofollow links had a similar statistical relationship with domain mentions in AI answers. That relationship could reflect visibility, referral exposure, source prominence, or other factors associated with the sites linking to a brand.

The image-link finding also needs restraint. Image links had stronger correlations than text links in this particular analysis, but that does not establish that image links are generally more influential in AI citation systems. Image links may be associated with publishers, media coverage, or authoritative sites that create the underlying relationship. The result should be treated as a narrow observation, not a universal GEO rule.

Why the Semrush results are not causal evidence

The sample was limited to 1,000 domains, the metrics were proprietary, and the study did not establish that changing a backlink profile would produce a predictable change in AI visibility. It also measured domain-level relationships, whereas an AI assistant may cite a specific page because of page-level relevance or evidence.

Use the study to support a practical principle—evaluate referring-domain quality, topical relevance, and editorial context rather than link totals—not to justify a fixed number of links or a guaranteed citation outcome.

Are brand mentions without links important?

Brand mentions may be useful indicators of AI visibility, even when they do not contain clickable links. A mention in a relevant article, video transcript, product comparison, community discussion, or expert resource can contribute to the broader information environment in which a brand is recognized.

However, a brand mention and a linked citation answer different marketing questions:

An Ahrefs study published on December 12, 2025, analyzed approximately 75,000 brands. The sample was constructed by filtering for domains with a Domain Rating above 40 and a highest-volume keyword with at least 800 monthly searches. Ahrefs then analyzed millions of AI responses using Brand Radar and used Spearman correlations. This is a large proprietary sample, but it is not a random sample of every business: the filters favor established domains and may underrepresent newer or smaller brands. (ahrefs.com)

The reported outcome was brand mentions in AI responses, not linked citations, impressions, or traffic. Ahrefs found that branded web mentions correlated more strongly with AI visibility than backlink counts or URL Rating in the studied surfaces. Its reported Domain Rating correlations with AI mentions were:

• ChatGPT: 0.266
• Google AI Mode: 0.285
• Google AI Overviews: 0.326

Branded-anchor correlations were:

• ChatGPT: 0.511
• Google AI Mode: 0.628
• Google AI Overviews: 0.527

These are correlations with brand appearances, not evidence that branded links cause AI assistants to cite a company. The stronger relationship for Google AI Mode may reflect that platform’s prompt mix, source systems, or tendency to surface established brands. It should not be assumed to apply to Gemini, ChatGPT Search, or Perplexity without platform-specific evidence.

What does the evidence say about Gemini?

Gemini should not be grouped with every other assistant as if the systems make citation decisions in the same way.

Semrush included Gemini in its five-platform sample and reported that Gemini showed a slightly stronger relationship with nofollow links than some other platforms, while Google AI Overviews and Perplexity leaned more toward follow links. Those observations came from the Semrush correlation analysis and do not demonstrate that Gemini uses nofollow links as a ranking signal. (semrush.com)

The Ahrefs study supplied the most prominent platform-specific coefficients for ChatGPT, Google AI Mode, and Google AI Overviews. Its methodology also references a shared question pool that includes Gemini, Copilot, and Perplexity, but the published headline findings discussed above should not be generalized to Gemini unless the relevant Gemini results are reported separately. (ahrefs.com)

For marketers, the practical lesson is simple: track Gemini independently. Do not assume that a tactic associated with Google AI Overviews or ChatGPT will produce the same result in Gemini.

Can strong content overcome a weaker backlink profile?

Sometimes. A page with fewer links may still be retrieved when it directly answers the prompt, provides verifiable evidence, and is easy for a system to interpret. That is a reasonable working hypothesis, not a guarantee.

The 2024 GEO research paper introduced GEO-bench and tested content-optimization methods in generative-engine responses. Its abstract reports visibility improvements of up to 40%, but the result refers to the paper’s experimental benchmark and optimization framework—not to a general increase in linked citations across ChatGPT, Google, Gemini, and Perplexity. The paper also states that results varied by domain. (arxiv.org)

The study should therefore be described precisely:

• It evaluated visibility in generative-engine responses, not universal citation performance.
• The result came from controlled experiments using GEO-bench, not a long-term marketing campaign.
• The reported maximum was up to 40%, not an expected average improvement.
• The findings do not establish that citations, quotations, or statistics will improve AI citations across every platform or topic.

For content teams, evidence-rich writing remains a sensible priority. Put the direct answer near the top, define important terms, cite primary or authoritative sources, distinguish facts from opinions, and make claims specific enough to verify. These practices improve usefulness and may improve retrieval or source selection, but they do not bypass platform-specific systems.

Why do studies disagree about backlinks and AI citations?

The studies measure different populations, platforms, metrics, and units of analysis.

Surfer reported near-zero overall Spearman correlations between citation frequency and three authority metrics: PageRank -0.065, Harmonic Centrality 0.010, and Surfer Domain Score -0.051. Platform-level values also remained close to zero. The analysis covered AI Mode, AI Overviews, ChatGPT, and Perplexity, but it did not measure every possible off-page signal or prove that authority is irrelevant in all retrieval systems. (surferseo.com)

The apparent conflict with Semrush is not necessarily a contradiction. Semrush measured domain mentions across a selected set of domains; Surfer measured the likelihood that source URLs were cited across a prompt dataset. A domain can be frequently mentioned without a particular page being selected as a citation. Differences in prompt sampling, platform versions, proprietary metrics, selection bias, and volatility can also change the observed relationships.

It is reasonable to infer that AI systems may respond to multiple signals of prominence, relevance, and source usefulness across the web. That mechanism is an inference, not an established universal rule disclosed by all AI providers.

What should marketers do differently from conventional link building?

Backlink building still matters when it earns genuine attention and supports normal SEO. GEO changes the objective: do not optimize only for link equity. Optimize for relevant discoverability, entity recognition, evidence, and measurable appearance in AI answers.

Prioritize the work in this order:

1. Earn relevant editorial links

Target sources that already influence your audience or cover your subject:

• Industry publications and trade associations
• Independent reviews and comparison sites
• Universities, research organizations, and professional bodies
• Reputable news outlets and specialist newsletters
• Relevant partner, supplier, and customer resources
• Expert communities where genuine participation is permitted

Evaluate each referring domain for topical relevance, editorial standards, audience fit, organic visibility, and whether the page is actually useful to readers. Use natural anchors, including branded or descriptive phrases where they accurately describe the destination. Avoid paid links intended to manipulate rankings, large-scale link exchanges, automated placements, and other schemes that violate search-engine spam policies.

2. Build entity and brand mentions

Pursue accurate references in places where people research your category, including reviews, expert roundups, podcasts, videos, forums, comparison pages, and industry documentation. A mention without a link may still be useful for recognition, but it should be earned through real expertise or public relevance rather than manufactured at scale.

3. Publish evidence-rich answer pages

Create pages for specific customer questions. Lead with a direct answer, include definitions and limitations, cite primary sources, show methodology for original data, and keep important facts in readable text. Use structured data where it accurately describes the page, but do not treat schema as an AI citation guarantee.

4. Improve crawlability and indexing

Check robots directives, canonicalization, rendering, internal links, XML sitemaps, duplicate pages, server reliability, and indexing status. For Google’s generative AI features, indexed pages that are eligible for normal Search snippets are the relevant baseline; inclusion is still not guaranteed. (developers.google.com)

5. Track mentions separately from linked citations

For each monitored prompt, record the platform, date, question, brand mention, cited domain, cited URL, competitors, sentiment, and topic. Report brand visibility and linked citations as separate metrics. Also compare the quality of referring domains rather than celebrating increases in total backlinks.

Tools such as LazySEO can support this workflow by helping teams monitor AI mentions, citations, competitors, and content gaps. The tool should complement—not replace—editorial outreach, technical SEO, and fact-checked content development.

FAQ

Do nofollow links help AI visibility?

They may be associated with AI visibility, but the evidence is limited and correlational. In Semrush’s 1,000-domain study, follow links had correlations of 0.334 Pearson and 0.504 Spearman with AI mentions, while nofollow links had 0.340 Pearson and 0.509 Spearman. This does not prove that nofollow links pass conventional ranking equity or directly cause citations. (semrush.com)

Can a brand mention without a link improve AI visibility?

It may help a brand become more recognizable in relevant web contexts, but current studies do not establish a universal causal mechanism. Track unlinked mentions separately from linked citations because they measure different outcomes.

Is link quantity or link quality more important?

Quality is the safer priority. Seek relevant, editorially earned links from credible referring domains with a real audience. Do not use backlink totals as a standalone AI-visibility target.

Do backlink effects differ by platform?

Yes. Semrush included ChatGPT, ChatGPT with Search, Gemini, Google AI Overviews, and Perplexity, while Ahrefs reported detailed correlations for ChatGPT, Google AI Mode, and Google AI Overviews. Differences in retrieval systems, prompts, indexes, and platform updates mean that results should be monitored separately rather than generalized. (semrush.com)

Does link building still matter for GEO?

Yes, when it earns relevant editorial coverage, supports discovery, strengthens brand recognition, or improves conventional search performance. It should not be treated as a direct control over AI citations. Pair link earning with brand mentions, evidence-rich pages, strong technical foundations, and platform-specific measurement.

What is the most reliable AI citation metric?

There is no single metric that reliably predicts citations across all assistants. Track actual prompt-level outcomes—brand mentions, linked domains, cited URLs, competitors, and citation frequency—then analyze them by platform, topic, and date.

Key takeaway

Backlinks are supporting signals, not direct citation switches. Earn fewer but more relevant editorial links, expand accurate brand mentions, publish well-supported answer pages, keep them crawlable and indexed, and measure AI mentions separately from linked citations. No current backlink metric reliably predicts citation likelihood across all AI assistants.

References

• https://developers.google.com/search/docs/appearance/ai-features?kgs=aa0bcc3d152ed142
• https://developers.google.com/search/docs/appearance/ai-features
• https://ahrefs.com/blog/ai-citations-vs-impressions-study